Skip to main content

Controlled change for GMP AI

Treat AI workflow change as a quality decision

Changes to models, prompts, retrieval, rules, tools, and workflows can alter behaviour and evidence. A proportionate AI governance change path keeps learning visible and reviewable.

Discuss your GMP AI pathway

Change is more than a code deployment

In a governed AI workflow, behaviour can change when the model, prompt, retrieval sources, rules, tools, thresholds, or surrounding process changes. The team needs a way to decide what the change means for quality and evidence.

  • Identify the change

    Describe what changed and which process, output, control, or evidence path may be affected.

  • Assess impact

    Consider intended use, output class, review requirements, data integrity, and validated-state implications.

  • Decide the response

    Define approval, evaluation, rollback, monitoring, or revalidation actions proportionate to the risk.

Changes worth assessing explicitly

  • Model or deployment changes

    A new model version or runtime can change output behaviour and evaluation results.

  • Prompt and policy changes

    Instructions, policies, and guardrails can alter what the workflow produces or permits.

  • Retrieval and reference changes

    New, removed, or changed source material can affect evidence, recommendations, and reviewer context.

  • Tool and workflow changes

    New permissions, actions, thresholds, retries, or handoffs can change process impact.

A controlled response

  • Classify the change

    Use impact and output boundary to decide the review depth.

  • Evaluate and approve

    Run the defined checks, record the decision, and retain the relevant evidence.

  • Monitor after release

    Watch quality, exceptions, review outcomes, and escalation signals after the change.

Next steps

Continue the decision

Have a change to assess?

Bring the model, prompt, retrieval, tool, or workflow change to a practical governance discussion.

Discuss your pathway